AgentSkillsCN

validate-data

在实现钩子、查询、边缘函数，或 RLS 策略之后，务必对功能数据层与真实数据库内容进行核对。当您在实现数据依赖型功能，或在调试空值/缺失数据时，自动触发该技能。

SKILL.md
--- frontmatter
name: validate-data
description: "Verify feature data layer against real database content. Use after implementing hooks, queries, edge functions, or RLS policies. Auto-invoke when implementing data-dependent features or when debugging empty/missing data."

Real Data Validation Skill

After implementing a data layer (hooks, queries, edge functions, RLS policies), run this verification BEFORE building UI components on top of it.

Step 1: Identify Expected Data

  • •What records should this feature return for existing users/entities?
  • •What tables/views are being queried?
  • •What joins, filters, or RLS policies are applied?

Step 2: Query Real Data

Use Supabase MCP (if available) or write a quick verification script to:

  • •Query the target table(s) directly, bypassing the application layer
  • •Confirm that at least 2-3 real records exist that match the feature's criteria
  • •If querying via the application's hooks/functions, compare results to direct database queries

Step 3: Verify Access

  • •Test that RLS policies allow the intended access for the relevant user roles
  • •Confirm that joins resolve correctly (no silent failures from missing foreign keys)
  • •Check that triggers produce valid, non-empty data for downstream consumers

Step 4: Diagnose or Proceed

If results are empty when data should exist:

  • •Check RLS policies first (most common cause)
  • •Check join conditions (are foreign keys populated?)
  • •Check filter logic (is matching too strict?)
  • •Check if backfill is needed for existing data
  • •FIX the data layer issue before building any UI

If results are correct:

  • •Proceed to UI implementation with confidence
  • •Note the verified query in the implementation plan for future reference